Why All My Agent Matrix Rooms Are E2EE by Default
If machines are going to talk to each other at 3am about your infrastructure, those transcripts are just as sensitive as your Slack channels.
AI Agents
4posts
Event-Driven Spokes vs. Polling: How I Cut Agent Latency 40×
HackerNews' Firebase SSE pushes new IDs in under a second. Our old cron polled every 5 min. Switching saved more than latency.
Cron Jobs Are the Missing Primitive for AI Agents
Most agent frameworks model agents as interactive. The useful work - watching, triaging, reporting - is scheduled.
LLM Serialization with fcntl: a 40-line Pattern for Single-Slot Inference Servers
Running llama.cpp with --parallel 1 means concurrent agents starve each other at the HTTP queue. A file lock solves it cleanly.